PreprintPDF Available

Politics of Artificial Intelligence Adoption: Unpacking the Regime Type Debate

Preprints and early-stage research may not have been peer reviewed yet.


What determines whether a country imports high-technology artificial intelligence (A.I.) products from the United States and China? Over the last decade, a growing body of literature began focusing on regime types and argue that authoritarian countries tend to import A.I. from China, whereas democratic countries from the United States. In this study we test this regime type hypothesis and find that both the US and China export to authoritarian and democratic countries alike, with China exporting to a larger number of countries. By employing multinomial logit with three leading A.I. trade datasets from Stanford University, Carnegie Endowment and World Bank, we find that Chinese A.I. exports are not driven by regime type, whereas American exports are. US A.I. exports are geared towards a smaller market of wealthy nations, whereas China supply to a broader market that is made up of a variety of regime types and GDP levels. Furthermore, we find that more countries import surveillance and policing-related A.I. products from the United States, compared to China, debunking a common mainstream view. We conclude by arguing military spending and GDP per capita are two of the strongest determinants of US A.I. exports, while a robust export pattern doesn’t emerge with Chinese A.I.
Politics of Artificial Intelligence Adoption:
Unpacking the Regime Type Debate
H. Akın ¨
Unver , Arhan S. Ertan
What determines whether a country imports high-technology artificial intelligence
(A.I.) products from the United States and China? Over the last decade, a growing
body of literature began focusing on regime types and argue that authoritarian countries
tend to import A.I. from China, whereas democratic countries from the United States.
In this study we test this regime type hypothesis and find that both the US and
China export to authoritarian and democratic countries alike, with China exporting
to a larger number of countries. By employing multinomial logit with three leading
A.I. trade datasets from Stanford University, Carnegie Endowment and World Bank,
we find that Chinese A.I. exports are not driven by regime type, whereas American
exports are. US A.I. exports are geared towards a smaller market of wealthy nations,
whereas China supply to a broader market that is made up of a variety of regime types
and GDP levels. Furthermore, we find that more countries import surveillance and
policing-related A.I. products from the United States, compared to China, debunking
a common mainstream view. We conclude by arguing military spending and GDP per
capita are two of the strongest determinants of US A.I. exports, while a robust export
pattern doesn’t emerge with Chinese A.I.
Keywords: artificial intelligence, political regimes, development, democracy,
Suggested Citation:
Unver, A and Ertan, A (2021). ”Politics of Artificial Intelligence Adoption: Unpacking
the Regime Type Debate”, in Filimowicz, M (editor), Democratic Frontiers: Algorithms
and Society, Routledge
Department of International Relations, Ozyegin University, Turkey,,
(ORCID: 0000-0002-6932-8325)
Department of International Trade, Bogazici University, Turkey,,
(ORCID: 0000-0001-9730-8391)
1 Introduction
Over the last decade, the diffusion and adoption of artificial intelligence (A.I.) and its re-
lated technologies across the world became a new flash- point of international competition.
Although developed Western nations were initially better positioned to seize primacy in this
new technological domain due to their stronger industrial base, the rapid advance and export
strategies of Chinese high-tech industries have facilitated the diffusion of A.I. into countries
with relatively less developed industrial and scientific foundations such as Ethiopia, Myan-
mar, and Zimbabwe. Such medium-cost, medium-capability A.I. exports so far included
speech and face recognition, personal data push services, interactive interface technologies,
data transmission infrastructure, and hardware devices required to transfer and process such
data (Gravett,2020). Often, these individual software and hardware bundles are exported
under ‘smart/ safe city’ or ‘smart policing’ marketing strategies, although the fundamental
‘detection-processing-prediction’ task chain is near-identical and re-deployable across various
projects (Roberts et al.,2021).
Chinese exports render automation at scale more affordable for developing countries by
trading readily deployable A.I. tools cheaper, and in most cases, assist automation efforts
of technologically less developed countries by reducing their dependence on trained human
capital. This cost-effective Chinese A.I. export doctrine helps importer countries stream-
line decisions without having to train and employ armies of highly skilled programmers or
engineers. While these countries cannot compete with more advanced and indigenously de-
veloped A.I. technologies of the US, EU, or Japan, they gain a clear advantage over their
regional rivals that cannot afford Western A.I. technologies, yet also don’t import from China
due to their US alliance commitments (Aly,2020). Over time, developing nations that em-
bargo Chinese A.I. exports due to their alliance commitments to the US or the EU or due to
ethical concerns about Chinese A.I. lose their competing power against other rivals that are
bound by neither and look to Beijing for high-tech imports. Therein lies the Chinese com-
parative advantage in A.I. exports and the core of the global competition for technological
influence between the US and China: Chinese strategy is as much about controlling global
cost-effective A.I. demand and defining the mainstream market as it is about competing with
the US.
As the technique and skillset ecosystem of A.I. expands, it becomes a general-purpose
technology (GPT). The more the A.I. ecosystem grows and its constituent technologies be-
come more complicated, what specifically such technologies entail becomes subject to greater
debate. While the general consensus of A.I. definition still focuses on human-like behavior
or decisions from machines, the scale and accuracy at which such expressions manifest form
the basis of the international competition over algorithms, talent, and hardware that render
these possible (Horowitz,2018). In simpler terms, ‘A.I. exports’ are generally defined as
hardware, algorithm and software units, bundles, and combinations that are intended to
automate a large array of tasks that are repetitive in nature, but require human-like vision,
decisions, and assessment procedures (Greitens,2020;Helm et al.,2020). These mostly
relate, but aren’t exclusive to, a combination of advances in machine learning, robotic and
vehicular autonomy, large-scale complex statistics, human-machine interaction, computer vi-
sion/language/agents, and neural networks. As a developing ecosystem of technologies, what
specifically A.I. can do at the strategic level is still being hypothesized at the military, so-
cial, economic, healthcare, and educational domains with some applied convergence (Wang,
2019). At the very least, grasping the true impact of A.I. on broader global economic and
social systems will take at least another decade, given how long it took for previous GPTs
to be absorbed into their zeitgeist (Petralia,2021). With such a lack of generally accepted
parameters over what specifically ‘A.I. technologies’ entail and are expected to do, countries
find it increasingly difficult to assess whether their national A.I. strategies are successful or
how to strategize wielding A.I. in trade, defense, and diplomacy. This is why, while most
countries realize that investing in A.I. is important, they are largely divided over how much
they should invest in it (or which specific component) and which leading nations to cooperate
with to maximize the chances of a successful A.I. adoption (Meltzer and Kerry,2021).
Amidst this uncertainty, the US and China grew locked into a competition over global
A.I. dominance, both trying to maximize their respective capacities and exporting their
own products to the world. Both countries prioritize capitalizing on the rapid growth in
computing capacity, producing and processing increasingly bigger datasets; develop newer
and more sophisticated algorithms and statistical methods to increase the quality of their
predictions, and creating an investment ecosystem that can financially sustain these advances
over the long term. Additionally, both countries seek to leverage A.I. to build and maintain
new alliance and partnership patterns that can bolster their international standing and
export dominance. Quite often, this political feud spills over into complementary technologies
such as 5G cellular networks, quantum computing, semiconductors, and autonomous vehicles,
broadening the trenches of the competition over emerging technologies, and rendering this
competition a truly global one with system-level balance of power repercussions Horowitz
et al. (2018).
2 Politics of Artificial Intelligence: The Regime Type
In the last few years, scholarship on how countries adopt and employ A.I. tools and tech-
niques grew significantly. Amidst this debate, major fields of inquiry sought to explain the
spread of such technologies from a diverse array of theories such as development and growth,
organizational adaptation, welfare and inequality, politics, and international competition
(Alsheibani et al.,2018;Weber and Sch¨utte,2019;Webster and Ivanov,2020;Zerfass et al.,
2020). Within the political science and international relations scholarship, researchers have
largely focused on whether governance and political systems affect how countries adopt new
technologies. In other words: does regime type (whether a country is democratic or author-
itarian) affect whether or how countries adopt artificial intelligence (Levy,2018;Malmborg
and Trondal,2021;Schiff et al.,2021;Unver,2018)? As promising extensions of this research
question, further inquiries emerged into whether automation at scale will be more conducive
to democracy or authoritarianism or whether unemployment and labor shifts generated by
mass automation will reinforce autocratic tendencies of governments (Feldstein,2019b;Frank
et al.,2019;Zeng,2020).
Since technology adoption is a multilayered process that is determined at the intersection
of a nation’s government, citizens, and corporations, the political system under which these
actors operate determines the scope and depth of such adoption (Evans,1995). In the
case of A.I., technology companies that drive high-tech adoption rely on sound regulatory
environment, uninterrupted and predictable funding streams, and an investment environment
that is conducive for international collaborations and partnerships (Milner,2006). In line
with the adoption patterns of other digital technologies, when a political system is free
enough to secure a rapid development of private technological enterprise and enable citizens
and companies to use new products produced by such companies unhindered, one can expect
democratic countries to lead in A.I (Corrales and Westhoff,2006).
Governments have a number of incentives in providing a conductive environment for A.I.
technology adoption. Automation at scale renders a wide array of tasks in national defense,
industrial production, banking/finance, and communication faster at higher volumes and in-
creasingly more accurate (Diamond,2010;Viscusi et al.,2020). A robust A.I. infrastructure
itself positions nations better to adopt rapidly developing advanced technologies and thereby
increase the competitiveness of those countries’ defense, commerce, industry, and trade sec-
tors (Haner and Garcia,2019). To that end, willingness to adopt and develop advanced A.I.
know-how generates global interest irrespective of regime types.
Yet, there have been studies that emphasize particular characteristics of democracies to
distinctly enable a faster and more robust adoption of newer technologies. According to this
stream, freer political systems and less centralized economic systems are better positioned to
create scientific and industrial ecosystems that can enable better adoption of digital technolo-
gies (Kimber,1991;Leslie et al.,2021;Nemitz,2018;Wright,2020). Corrales and Westhoff
(2006), for example, demonstrate that democracies, on average, adopted the Internet faster
and with greater nationwide penetration compared to authoritarian governments. Milner
(2006), however, posits that democracies are not inherently more conducive to the adoption
of new technologies, but elite preferences in autocracies define whether that country will
adopt an emerging technology. This line of argumentation suggests that if authoritarian
leaders believe a new technology will reinforce their rule, they will create conducive funding
and investment environment to facilitate new technology adoption. For example, an author-
itarian country may restrict the adoption and spread of the Internet and social media due to
their regime-weakening effects, but actively facilitate and encourage the development of an
A.I. industry due to their defense and surveillance implications. In Acemoglu and Robinson
(2000)’s words: ‘agents who have political power and fear losing it who will have incentives
to block technological progress’.
However, Stier (2015) introduces a temporal argument in his cross-sectional study, em-
pirically demonstrating that after 2013, there are no clear differences in Internet adoption
between democracies and authoritarian countries, as the latter have caught up rapidly once
they learned how to manage and control digital communication outlets (Druzin and Gordon,
2018;Rød and Weidmann,2015). In that vein, technology adoption can also be viewed as a
matter of regime survival, whereby active elite preferences matter more than passive regime
type effects. Once such technologies demonstrably reduce costs of surveillance and control,
they trigger a very different form of technology diffusion across hybrid and authoritarian
regimes that cannot be explained by the democratic technology diffusion literature (Choi
and Jee,2021). Kania (2021)’s exploration of Chinese People’s Liberation Army doctrine
on artificial intelligence is particularly important in this sense, as it is expected to influ-
ence defense doctrines of other developing nations that are likely to import A.I. components
heavily from China. Kania (2021) demonstrates how China has situated its armed forces
and revisionist territorial claims in the South China Sea as the main engine of its broader
A.I. efforts, outlining that authoritarian countries with similar revisionist claims may find it
more desirable to acquire advanced technologies from China rather than Western nations.
But why would A.I. adoption patterns flow across similar regime types? One line of schol-
arship explores whether A.I. is a distinctly liberal or illiberal technology. Wright (2020), for
example, focuses on the surveillance-enhancing aspects of A.I. to posit that such technolo-
gies, under authoritarian hands, will reinforce further repression and will embolden other
authoritarian countries to adopt A.I. tools that make it easier to engage in large-scale spying
on citizens. To prevent such a scenario, he emphasizes the role of civil society and democratic
institutions to bolster international A.I. governance norms to facilitate technology diffusion
among democratic nations, thereby rendering authoritarian adoption of A.I. undesirable for
developing nations. Lamensch (2021) posits a similar argument, demonstrating how A.I. in
particular and digital technologies in general are more conducive for authoritarian tenden-
cies rather than liberal ones by reducing the costs of large-scale surveillance. That said, she
underlines the fact that a broad range of A.I. tools used by authoritarian regimes are de-
veloped and exported by democracies themselves, blurring the lines between various regime
types and how much they contribute to global authoritarianism.
Polyakova and Meserole (2019), however, argue that it is no longer democracies alone that
are exporting A.I. technologies to authoritarian countries, but Chinese and Russian low-cost
options are increasingly clawing back market share from Western exports, creating distinct
regime type cleavages in technological trade patterns. They argue in favor of stronger export
controls to minimize the transfer of surveillance and A.I.-related technologies from democ-
racies into authoritarian regimes and opt for targeted sanctions on critical technologies in
more extreme cases. Similarly, by deploying a large dataset of A.I. adoption trends,Feldstein
(2019b) outlines three main pathways for how A.I. can reinforce illiberalism: mass surveil-
lance that disables popular movements and protests, localized surveillance to suppress dissent
in protest-prone regions, and deploying organized large-scale disinformation to delegitimize
the political opposition. He posits that while these techniques are problematic enough under
Chinese control, exporting such technologies to bolster autocracies across the world reduces
developing countries’ reliance on Western A.I. exports, thereby bypassing the need to adopt
democratic norms and practices as prerequisites to access advanced technologies.
As an increasing number of democratic regimes began importing A.I. tools from China,
the regime type became particularly salient from an international alliances point of view –
especially on whether A.I. trading patterns will generate new alliance formations such as
hybrid ones that include a mixture of regime types. Since post-World War II and post-Cold
War world orders are built on institutionalizing cooperation among democracies and forming
a unified bloc against authoritarian countries, A.I.-related conflict and cooperation patterns
are generally expected to follow a similar course. Franke (2021), for example, situates the
EU firmly within the US-led A.I. alliance system, arguing that regime type and democratic
norms will remain as the fundamental conditions for cooperation and partnership patterns
in the next decade. Additionally, she prescribes that to attain primacy in the new A.I.
competition, the EU would have to prioritize bolstering the private sector and technology
startups rather than state-led initiatives by individual member states. Horowitz (2018)
argues that international alliance and balance of power shifts will largely be dictated by
whichever country leads in the scientific and military advances in A.I. He warns that a US-
led A.I. supremacy is by no means guaranteed, and if the current trend in deep learning
research continues, it might be likelier for China to create a new partnership regime – one
that may even include most European countries – based on its lower-cost high technology
As outlined in Johnson (2019), an increasing number of countries view A.I. as a strategic
investment priority and view any progress in this field as a national security goal. As a
testament to A.I.’s growing importance, while there were only three countries with national
A.I. strategies in 2016, as of 2021, there are more than 60. In one of the most comprehensive
of such analyses, Fatima et al. (2020) find great discrepancies over how countries discur-
sively construct A.I. in their national strategy documents. They conclude that democratic
countries with a lower technology base are more likely to emphasize ethical dimensions of
A.I. in their national strategy documents compared to democratic or authoritarian/hybrid
regimes with a stronger technology base. Yet, even countries with no national strategy doc-
uments are involved in strategic A.I. trade. Campbell (2008) focused on China’s ex- ports
to sub-Saharan African countries underlining that Beijing was reinforcing existing illiberal
tendencies of countries by rendering repression easier. Zeng (2020) advances this argument
further, positing that the fundamental Chinese A.I. strategy is to bolster states against soci-
eties and thereby attract surveillance-oriented developing countries into Chinese technology
exports orbit.
From this survey, we arrive at the following hypotheses:
H1: Authoritarian regimes are more likely to choose Chinese A.I. exports
H2: Less developed and poorer countries are more likely to choose Chinese A.I. exports
H3: Countries that have existing strong trade relations with China are more likely to
acquire A.I. technologies from them
H4: Authoritarian countries are more likely to acquire surveillance and repression-
related A.I. technologies
H5: Belt and Road Initiative (BRI) countries are more likely to acquire Chinese A.I.
3 Data Sources and Methodology
In order to test these hypotheses, we combined two of the most comprehensive datasets that
dissect A.I. trade and adoption patterns across the world. The first of these datasets is the
Artificial Intelligence Index1 published by Stanford University’s Human-Centered Artificial
Intelligence (HAI), which logs A.I. adoption trends across seven categories: research and
development, technical performance, economy, A.I. education, ethical challenges of A.I.,
diversity in A.I., and A.I. policy and national strategy. These categories are then merged
into the ‘Global A.I. Vibrancy Tool’, which measures countries’ A.I.-related progress across
22 indicators, including conference papers, journal citations, patents, funded companies,
and skill penetration. The second dataset we use is the A.I. Global Surveillance (AIGS)
Index, published by Carnegie Endowment, which combines regime type scores, military
expenditures, major tech firms in operation, and the type of A.I. technology used by 73
countries (Feldstein,2019a). The dataset also contains information whether China, US,
or Japan is actively involved in those countries’ A.I. ecosystems. AIGS also uses Freedom
House weighted average country scores, Economist Intelligence Unit (EIU) Democracy Index
2018 and V-Dem’s Electoral Democracy Index, to measure regime type as well as Stockholm
International Peace Research Institute’s (SIPRI’s) military expenditure dataset to measure
budget and spending-related figures. BRI country information comes from Kliman and Grace
(2018), and Chinese overseas direct investment figures are compiled from EIU. Our trade
(export-import) data comes from World Bank’s World Integrated Trade Solution (WITS)
In order to analyze our hypotheses, we estimate regression models in the following linear
where Yis one of our dependent variables, Xis a vector of explanatory and Zis a vector
of control variables in our dataset 2and ϵis the random error term. As for the estimation
method, we have utilized the multinomial logit since our dependent variable has more than
two mutually exclusive and exhaustive categories which are nominal in nature and do not
have a meaningful sequential order. Multinomial logit models are a direct generalization
of the ordinary two-outcome (binary) logit models and are used to estimate relationships
between an ordinal (categorical or ordered) dependent variable and a set of independent
variables. In multinomial logit, an underlying score is estimated as a linear function of the
independent variables and a set of cut-points and the error term is assumed to be logistically
distributed. Parameter estimation is performed through an iterative maximum likelihood
algorithm (Hausman and McFadden,1984;Small and Hsiao,1985).
4 Findings
Our results demonstrate a more nuanced picture with regard to the extent regime type influ-
ences whether countries prefer American or Chinese A.I. exports (Hypothesis-1). Primarily,
we offer a more inequality- and development-oriented explanation (i.e., gross domestic prod-
uct [GDP] differences between countries and whether a country is industrialized and has an
existing strong technological base) rather than a regime type explanation (whether a country
is democratic or authoritarian) over states’ A.I.-related acquisition choices.
2See Table 1 for the detailed list of these variables
With regard to our Hypothesis-2, countries with a greater number of existing A.I. firms
tend to rely on both American and Chinese A.I. investments, whereas countries with slightly
fewer number of A.I. firms rely predominantly on the US. However, countries that have few
or no A.I. firms import mainly from China (see Figure 1 and Table 1). This is an interesting
finding, since the number of existing firms in a country and existing technological base push
that technology ecosystem towards both the US and China rather than just the US alone.
This means that top performer countries in A.I. diversify their technology imports between
two technology superpowers and do not rely wholly on the US alone. However, countries
with higher A.I. investment budgets prefer US A.I. exports, whereas countries with lower
A.I. investment budgets import heavily from China, producing a clear inequality hypothesis
(see Figure 2). Therefore, greater existing technological base, if measured by the number of
existing A.I. firms, pushes countries to diversify into US and Chinese exports. However, if
existing technological base is measured by countries’ investment budgets, then such countries
tend to rely primarily on the US.
As far as Hypothesis-3 is concerned, existing trade relations are strong determinants of
countries’ A.I. import choices. Countries where China covers a larger portion of imports than
the US tend to import A.I. technologies from China at a significant rate (see Figure 4 and
Table 1). Moreover, countries that have increasingly become more dependent on trade with
China in the last 20 years (growth of imports from China versus from the US between 2000
and 2019), strongly prefer Chinese A.I. exports to American ones (see Figure 5 and Table
1). This kind of clear difference doesn’t appear in countries where the EU is a dominant
trade partner; in countries where the EU is the largest trading partner, preference for A.I.
technology from China, US, or both show similar results (suggesting EU-origin A.I. export
preferences). Yet, in countries where trade with the EU has grown stronger over the last
20 years, preference for Chinese A.I. imports is significantly lower (see Figure 6). This
reinforces the hypothesis that Chinese A.I. exports fill the gap in countries where both
American and European trade have dwindled, and thereby situate China as an alternative
source of development funds in countries that cannot access any Western sources.
Exploring Hypothesis-5 yields additional insights that weaken the regime-type argument.
Being a ‘Belt and Road Initiative country’ (BRI Group) is a strong indicator of whether a
country imports such infrastructure from China (see Figure 3). However, even in non-BRI
countries, preference is towards importing both from China and the US rather than heavily
or solely from the US. This means that non-BRI countries choose both Chinese and American
A.I. exports rather than US-made alone. Therefore, although BRI is a strong anchor that ties
a wide range of countries to China’s global efforts, Chinese A.I. presence is strong in countries
that aren’t within the BRI framework. This somewhat refutes arguments of a ‘new A.I. Cold
War’ between the US and China (Garcia,2021), as in terms of global market dominance,
China is the leading supplier of A.I. technologies to a broader and more numerous range of
countries. In most countries the US export to, American A.I. firms have to compete with
their Chinese counterparts.
As for our Hypothesis-1 and Hypothesis-4, with regard to ‘authoritarian A.I. tools’ such
as facial recognition and surveillance exports, we see no clear difference between the US or
Chinese-origin import preferences: facial recognition and related technologies are imported
from either/both countries in a comparable fashion, weakening the US-led ‘democratic A.I.
exports’ argument (see Figure 7 and Figure 8). In terms of perhaps the ‘most authoritarian’
of A.I. exports – smart policing systems – importers have a distinct preference for US-made
exports or both US and Chinese offerings, further weakening the argument that China is the
only country that exports authoritarian tools or that the US leads ‘democratic A.I.’ efforts.
Further hammering in this point, a greater number of countries import surveillance-purpose
A.I. tools from the US compared to China. In terms of exporting A.I. as a repressive tool,
China and the US appear comparably involved.
Indeed, further delving into Hypothesis-1, we observe a weak relationship between free-
dom scores and whether countries prefer Chinese- or American-origin A.I. tools. Countries
with higher freedom scores very slightly tend to import US-made A.I. tools, but China is
a strong second, nullifying a great degree of regime type effects on A.I. imports. Countries
that are listed as ‘not free’, ‘partly free’, and ‘free’, all tend to import A.I. from China more
frequently, although the margin is larger with the former two (see Figure 14). Polyarchy
scores also show a weak relationship between political systems and A.I. import preferences
(see Figure 15). More interestingly, as regime scores improve, countries tend not to buy
either from the US or China (suggesting they buy either from the EU or Japan) or from
both the US and China. We further observe that electoral autocracies tend to buy more
from China, but this is similarly valid for electoral democracies as well as liberal democracies
(see Figure 9 and Figure 10 and Table 1). As far as regional variances go, Chinese A.I. ex-
ports dominate Africa and Asian continents, while they remain popular in Europe and Latin
America as well. It is important to underline that there are no European countries in our
dataset that are solely buying A.I. infrastructure from the US, but there are five European
countries buying solely from China (see Figure 13).
The best explanation of A.I. import choices remain GDP per capita and income (see Fig-
ure 11 and Table 1). Countries with low, lower-middle and upper-middle income buy dom-
inantly from China, while in high-income countries this gap narrows. Even then, richer
countries tend to buy from both China and the US rather than being dependent on either.
But most acutely, countries with a higher GDP per capita tend to buy more from the US
than China, whereas countries that have a higher GDP tend to buy from both countries.
Perhaps as a final nail in the coffin of the ‘regime type’ argument, countries that have a
higher military expenditure budget tend to buy A.I. exports from the US in a larger volume.
Countries with lower military budgets clearly prefer Chinese A.I. exports (see Figure 12).
5 Discussion
Overall, these results increase our skepticism towards the explanatory value of the ‘regime
type’ argument and show a weak relationship between whether a country is democratic
or authoritarian and whether it acquires A.I. technology predominantly from China or the
US. Rather, current evidence demonstrates that not only authoritarian regimes, but also
electoral and liberal democracies tend to import A.I. tools from China, often more exclusively
than they import from the US. With the exception of a few wealthy liberal democracies,
most countries – liberal democracies, electoral democracies, electoral autocracies, and closed
autocracies – all choose Chinese exports, and in most cases, countries with a richer ecosystem
of A.I. firms tend to import both from China and the US. With these findings, we argue
that regime type is not a strong determinant of A.I. acquisition trends, especially in light
of the fact that more countries buy smart policing and surveillance tools from the US than
they do from China.
Two of the most powerful variables that explain whether countries adopt A.I. through
China or the US are military spending and both the size and the per capita value of GDP.
Wealthier countries with larger defense budgets tend to prefer US- origin A.I. equipment
while China capitalizes on a larger A.I. market that includes all regime types except the
wealthiest liberal democracies. In addition, there is currently no strong evidence to support
the claim that countries import ‘authoritarian A.I. tools’ only from China and ‘democratic
tools’ only from the US. Most acutely, countries buy both authoritarian and otherwise A.I.
technologies from both China and the US, and in the case of surveillance-oriented tech-
nologies, the US has a distinct market edge. Furthermore, current data shows that global
competition for A.I. dominance is largely driven by global inequalities – between a smaller
group of wealthier countries that can afford to prioritize A.I. ethics and norms and a larger
group of developing countries that are seeking more affordable and faster options to integrate
emerging technologies into their national strategies.
These findings mean that the scholarship should move beyond the regime type dyad in as-
sessing why countries buy Chinese or American A.I. exports and instead focus on comparative
developmental explanations such as growth, regional trade competition, and affordability of
technological adoption. Overall, our findings indicate that China will continue to exclusively
trade with a larger number of countries, if it is the only country that is able to provide af-
fordable A.I. exports at scale. It is important to underline that this doesn’t mean that China
is ‘better’ or ‘stronger’ in A.I. compared to the US,3but merely that its A.I. export model
will likely help it stay ahead of the US in terms of the number of countries it trades with
exclusively (with no US competition). While important from a policy perspective; Western
‘A.I. ethics’ discourse will likely remain insignificant in the face of immediate growth and
technological advancement needs of developing countries. Given the popularity of Chinese
A.I. exports, the majority of countries out- side the US/EU alliance ecosystem will find
it easy to turn away from Western A.I. exports that are more expensive than developing
countries can reliably use, and come with ideological strings attached. Such turn towards
Chinese A.I. exports are even visible in the EU (related analyses / figures are available upon
request), where no country exclusively trades with the US on A.I. issues and opt for both
Chinese and American offerings.
To conclude, although US-origin A.I. exports may be technologically more advanced,
Chinese export strategy of offering more moderately priced, moderate-capability A.I. tools
has resulted in significant popularity of the latter globally. Unless either the US or EU can
produce a more imaginative economic technology-transfer model that can render high-tech
development more affordable for developing countries, Western A.I. exports will likely re-
main within the ‘rich countries club’ and will not be adopted by the rest of the world. If
that continues to be the case, China may well lead in A.I. in terms of the number of its
3For a comprehensive study on this matter, see Castro et al. (2019)
trade partners, which can ultimately lead to Chinese diplomatic and political gains in other
technology-related issues. In terms of how to alter this momentum, our suggestion is an
outlier against other mainstream suggestions that urge building better norms, more compre-
hensive ethical guidelines, establish stronger Western institutions to safeguard A.I. ethics,
or issue targeted sanctions against Chinese A.I. exports or countries that import such tools.
While these are important, none of those suggestions are likely to shift developing coun-
tries away from Chinese A.I. exports – rather, they will more likely exclude and antagonize
these countries, expediting China’s ability to form a unified international bloc of developing
countries on A.I. technology partnership.
Instead, we believe that the US and the EU have to build a new export model that
can compete with China at the medium-cost, medium- capability A.I. exports range. Since
Western exports are already unaffordable for the majority of the countries advancing in A.I.,
the only way to link those countries to the Western technology ecosystem is to offer more
reliable and better-performing, mid-range exports at scale. We do not suggest either the
US or Europe turn away from A.I. ethics or norms, but in order to be able to insist on
those priorities, Western exports need to be decisively more preferable over Chinese exports
at price-performance ratio. Ultimately, by breaking Western A.I. exports from its small
‘rich countries club’ bubble and by offering more efficient A.I. technologies can the West
produce sufficient leverage to build international partnerships based on truly representative
and inclusive global institutions and norms.
Figures and Table
Figure 1: Chinese versus US A.I. import preference in sample countries according to the
median number of existing A.I. firms
Figure 2: Chinese versus US A.I. import preference in sample countries according to the
median size of existing A.I. investment
Figure 3: Chinese versus US A.I. import preference in sample countries by Belt & Road
Initiative participation
Figure 4: Chinese versus US A.I. import preference in sample countries by ratio of imports
from China to from USA
Figure 5: Chinese versus US A.I. import preference in sample countries by the ratio of import
growth from China
Figure 6: Chinese versus US A.I. import preference in sample countries by the ratio of import
growth from the EU
Figure 7: Chinese versus US A.I. import preference in sample countries by surveillance-
related A.I. tools
Figure 8: Chinese versus US A.I. import preference in sample countries by smart policing-
related A.I.
Figure 9: Chinese versus US A.I. import preference in sample countries by regime type
Figure 10: Chinese versus US A.I. import preference in sample countries by freedom scores
Figure 11: Chinese versus US A.I. import preference in sample countries by GDP per capita
Figure 12: Chinese versus US A.I. import preference in sample countries by military spending
Figure 13: Chinese versus US A.I. import preference in sample countries by region
Figure 14: Chinese versus US A.I. import preference in sample countries by level of freedom
Figure 15: Chinese versus US A.I. import preference in sample countries by polyarchy scores
(1) (2) (3) (4) (5) (6) (7)
Equation VARIABLES AIsource
China_AI Regime = EA 1.7228 2.2589 2.1755 1.9412 1.7436 1.7436 1.0466
(1.3402) (1.6000) (1.5775) (1.5645) (1.5604) (1.5604) (1.8036)
Regime = ED -0.5596 -0.0121 0.0829 -0.0572 0.1264 0.1264 0.6703
(0.9777) (1.1033) (1.0915) (1.1308) (1.0801) (1.0801) (1.4308)
Regime = LD -0.6286 0.4927 1.2213 1.3516 1.2765 1.2765 1.7412
(1.1407) (1.4767) (1.5065) (1.6161) (1.5970) (1.5970) (1.6773)
LNgdppc_2019 -0.4777 -0.4777 -0.9035
(0.8051) (0.8051) (1.0229)
LNgdp_2019 0.4279 0.4279 0.7619
(0.5459) (0.5459) (0.5280)
CHN/US imp. rat.
CHN/US imp. gr. 2.7650*
EUR imp. growth -3.4151*
USA_AI Regime = EA 0.6931 1.3961 1.3518 1.3471 5.2275 5.2275 28.8147***
(1.8838) (2.0514) (2.0376) (2.0439) (3.5103) (3.5103) (3.2972)
Regime = ED -1.2528 -0.6505 -0.6495 -0.7278 3.4148 3.4148 -124.8285***
(1.6369) (2.1023) (2.1801) (2.2883) (3.8786) (3.8786) (4.2193)
Regime = LD 0.2877 1.9964 2.1330 2.1493 4.6416 4.6416 -4.4278
(1.5381) (1.7753) (1.8358) (1.8961) (3.5507) (3.5507) (3.1447)
LNgdppc_2019 2.2580* 2.2580* 104.3610***
(1.3289) (1.3289) (3.6983)
LNgdp_2019 0.8505 0.8505 -4.5430***
(1.1788) (1.1788) (0.7579)
CHN/US imp. rat.
CHN/US imp. gr. -438.0349***
EUR imp. growth 302.0774***
both Regime = EA 1.6094 1.9530 1.9871 1.2224 2.5591 2.5591 4.0372
(1.4935) (1.7387) (1.7161) (1.7664) (1.6592) (1.6592) (2.5536)
Regime = ED -0.0000 0.1111 -0.0686 -0.7744 0.1516 0.1516 -0.0096
(1.1417) (1.2616) (1.2614) (1.4057) (1.4816) (1.4816) (1.7183)
Regime = LD 0.9808 1.6360 1.2047 1.5110 1.7403 1.7403 2.7003
(1.2160) (1.5768) (1.7878) (1.9485) (1.9483) (1.9483) (2.2363)
LNgdppc_2019 1.4890 1.4890 1.4455
(0.9747) (0.9747) (1.1832)
LNgdp_2019 0.2503 0.2503 0.3407
(0.4185) (0.4185) (0.6122)
CHN/US imp. rat.
CHN/US imp. gr. 0.4983
EUR imp. growth 2.5039
Pseudo R2
Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1)
Included Controls: region, population, BRI
Acemoglu, D. and Robinson, J. A. (2000). Political Losers as a Barrier to Economic Devel-
opment. American Economic Review, 90(2):126–130.
Alsheibani, S., Cheung, Y., and Messom, C. (2018). Artificial Intelligence Adoption: AI-
readiness at Firm-Level. In PACIS 2018 Proceedings, 9.
Aly, H. (2020). Digital Transformation, Development and Productivity in Developing Coun-
tries: Is Artificial Intelligence a Curse or a Blessing? Review of Economics and Political
Science. Publisher: Emerald Publishing Limited.
Campbell, H. (2008). China in Africa: Challenging US global hegemony. Third World
Quarterly, 29(1):89–105. Publisher: Taylor & Francis.
Castro, D., McLaughlin, M., and Chivot, E. (2019). Who Is Winning the AI Race: China,
the EU or the United States? Technical report, Center for data innovation.
Choi, C. and Jee, S. H. (2021). Differential Effects of Information and Communication
Technology on (De-) Democratization of Authoritarian Regimes. International Studies
Quarterly, 65(4):1163–1175. Publisher: Oxford University Press.
Corrales, J. and Westhoff, F. (2006). Information Technology Adoption and Political
Regimes. International Studies Quarterly, 50(4):911–933. Publisher: Blackwell Publishing
Ltd Oxford, UK.
Diamond, L. (2010). Liberation Technology. Journal of Democracy, 21(3):69–83.
Druzin, B. and Gordon, G. S. (2018). Authoritarianism and the Internet. Law & Social
Inquiry, 43(4):1427–1457. Publisher: Cambridge University Press.
Evans, P. B. (1995). Embedded Autonomy: States and Industrial Transformation. Princeton
University Press.
Fatima, S., Desouza, K. C., and Dawson, G. S. (2020). National Strategic Artificial Intelli-
gence Plans: A Multi-dimensional Analysis. Economic Analysis and Policy, 67:178–194.
Feldstein, S. (2019b). The Road to Digital Unfreedom: How Artificial Intelligence Is Reshap-
ing Repression. Journal of Democracy, 30(1):40–52. Publisher: Johns Hopkins University
Feldstein, S. (September 17, 2019a). The Global Expansion of AI Surveillance. Carnegie
Endowment for International Peace.
Frank, M. R., Autor, D., Bessen, J. E., Brynjolfsson, E., Cebrian, M., Deming, D. J.,
Feldman, M., Groh, M., Lobo, J., Moro, E., Wang, D., Youn, H., and Rahwan, I. (2019).
Toward Understanding the Impact of Artificial Intelligence on Labor. Proceedings of
the National Academy of Sciences, 116(14):6531–6539. Publisher: National Academy of
Sciences eprint:
Franke, U. (2021). Artificial Intelligence Diplomacy: Artificial Intelligence Governance as
a New External Policy Tool. Technical report, European Parliament Directorate-General
for Internal Policies.
Garcia, D. (2021). Stop the emerging AI cold war. Nature, 593(7858):169.
Gravett, W. H. (2020). Digital Coloniser? China and Artificial Intelligence in Africa.
Survival, 62(6):153–178. Publisher: Routledge.
Greitens, S. C. (2020). Dealing with Demand for China’s Global Surveillance Exports.
Technical report, Brookings Institution.
Haner, J. and Garcia, D. (2019). The Artificial Intelligence Arms Race: Trends and World
Leaders in Autonomous Weapons Development. Global Policy, 10(3):331–337. Publisher:
Wiley Online Library.
Hausman, J. and McFadden, D. (1984). Specification Tests for the Multinomial Logit Model.
Econometrica, 52(5):1219–1240. Publisher: [Wiley, Econometric Society].
Helm, J. M., Swiergosz, A. M., Haeberle, H. S., Karnuta, J. M., Schaffer, J. L., Krebs, V. E.,
Spitzer, A. I., and Ramkumar, P. N. (2020). Machine Learning and Artificial Intelligence:
Definitions, Applications, and Future Directions. Current Reviews in Musculoskeletal
Medicine, 13(1):69–76.
Horowitz, M. C. (2018). Artificial Intelligence, International Competition, and the Balance
of Power. Texas national security review, 1(3). Publisher: Texas National Security Review.
Horowitz, M. C., Allen, G. C., Kania, E. B., and Scharre, P. (2018). Strategic Competition
in an Era of Artificial Intelligence. Center for a New American Security., p. 27 edition.
Johnson, J. (2019). Artificial Intelligence & Future Warfare: Implications for International
Security. Defense & Security Analysis, 35(2):147–169. Publisher: Routledge.
Kania, E. B. (2021). Artificial intelligence in China’s Revolution in Military Affairs. Journal
of Strategic Studies, 44(4):515–542. Publisher: Routledge.
Kimber, R. (1991). Artificial Intelligence and the Study of Democracy. Social Science
Computer Review, 9(3):381–398. Publisher: Sage Publications Sage CA: Thousand Oaks,
Kliman, D. and Grace, A. (2018). Power Play: Addressing China’s Belt and Road Strategy.
Technical report, Center for a New American Security.
Lamensch, M. (2021). Authoritarianism Has Been Reinvented for the Digital Age.
Leslie, D., Burr, C., Aitken, M., Cowls, J., Katell, M., and Briggs, M. (2021). Artificial
Intelligence, Human Rights, Democracy, and the Rule of Law: A Primer. Technical
Report SSRN Scholarly Paper ID 3817999, Social Science Research Network. Available at
SSRN 3817999.
Levy, F. (2018). Computers and Populism: Artificial Intelligence, Jobs, and Politics in the
Near Term. Oxford Review of Economic Policy, 34(3):393–417.
Malmborg, F. a. and Trondal, J. (2021). Discursive Framing and Organizational
Venues: Mechanisms of Artificial Intelligence Policy Adoption. International Review of
Administrative Sciences. Publisher: SAGE Publications Ltd.
Meltzer, J. P. and Kerry, C. F. (2021). Strengthening International Cooperation on Artificial
Intelligence. Technical report, Brookings Institution.
Milner, H. V. (2006). The Digital Divide: The Role of Political Institutions in Technology
Diffusion. Comparative Political Studies, 39(2):176–199. Publisher: SAGE Publications
Nemitz, P. (2018). Constitutional democracy and technology in the age of artificial intel-
ligence. Philosophical Transactions of the Royal Society A: Mathematical, Physical and
Engineering Sciences, 376(2133):20180089. Publisher: The Royal Society Publishing.
Petralia, S. (2021). GPTs and Growth: Evidence on the Technological Adoption of Elec-
trical and Electronic Technologies in the 1920s. European Review of Economic History,
Polyakova, A. and Meserole, C. (2019). Exporting Digital Authoritarianism: The Russian
And Chinese Models. Technical report, Brookings Institution.
Roberts, H., Cowls, J., Morley, J., Taddeo, M., Wang, V., and Floridi, L. (2021). The Chinese
Approach to Artificial Intelligence: An Analysis of Policy, Ethics, and Regulation. AI &
SOCIETY, 36(1):59–77. Publisher: Springer.
Rød, E. G. and Weidmann, N. B. (2015). Empowering Activists or Autocrats? The Internet
in Authoritarian Regimes. Journal of Peace Research, 52(3):338–351. Publisher: SAGE
Publications Ltd.
Schiff, D. S., Schiff, K. J., and Pierson, P. (2021). Assessing Public Value Failure in Govern-
ment Adoption of Artificial Intelligence. Public Administration. Publisher: Wiley Online
Small, K. A. and Hsiao, C. (1985). Multinomial Logit Specification Tests. International
Economic Review, 26(3):619–627. Publisher: [Economics Department of the University of
Pennsylvania, Wiley, Institute of Social and Economic Research, Osaka University].
Stier, S. (2015). Political Determinants of E-government Performance Revisited: Comparing
Democracies and Autocracies. Government Information Quarterly, 32(3):270–278.
Unver, A. (2018). Artificial Intelligence, Authoritarianism and the Future of Political Sys-
tems. Technical Report SSRN Scholarly Paper ID 3331635, Social Science Research Net-
Viscusi, G., Collins, A., and Florin, M.-V. (2020). Governments’ Strategic Stance toward
Artificial Intelligence: An Interpretive Display on Europe. In Proceedings of the 13th
International Conference on Theory and Practice of Electronic Governance, ICEGOV
2020, pages 44–53, New York, NY, USA. Association for Computing Machinery. event-
place: Athens, Greece.
Wang, P. (2019). On Defining Artificial Intelligence. Journal of Artificial General Intelligence,
Weber, F. D. and Sch¨utte, R. (2019). State-of-the-art and Adoption of Artificial Intelli-
gence in Retailing. Digital Policy, Regulation and Governance, 21(3):264–279. Publisher:
Emerald Publishing Limited.
Webster, C. and Ivanov, S. (2020). Robotics, Artificial Intelligence, and the Evolving Nature
of Work. In George, B. and Paul, J., editors, Digital Transformation in Business and
Society: Theory and Cases, pages 127–143. Springer International Publishing, Cham.
Wright, N. D. (2020). Artificial Intelligence and Democratic Norms. Technical report,
National Endowment for Democracy.
Zeng, J. (2020). Artificial Intelligence and China’s Authoritarian Governance. International
Affairs, 96(6):1441–1459.
Zerfass, A., Hagelstein, J., and Tench, R. (2020). Artificial Intelligence in Communication
Management: A Cross-national Study on Adoption and Knowledge, Impact, Challenges
and Risks. Journal of Communication Management, 24(4):377–389. Publisher: Emerald
Publishing Limited.
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
In the context of rising delegation of administrative discretion to advanced technologies, this study aims to quantitatively assess key public values that may be at risk when governments employ automated decision systems (ADS). Drawing on the public value failure framework coupled with experimental methodology, we address the need to measure and compare the salience of three such values—fairness, transparency, and human responsiveness. Based on a pre‐registered design, we administer a survey experiment to 1,460 American adults inspired by prominent ADS applications in child welfare and criminal justice. The results provide clear causal evidence that certain public value failures associated with artificial intelligence have significant negative impacts on citizens’ evaluations of government. We find substantial negative citizen reactions when fairness and transparency are not realized in the implementation of ADS. These results transcend both policy context and political ideology and persist even when respondents are not themselves personally impacted. This article is protected by copyright. All rights reserved. Available openly at:
Full-text available
China has adopted a proactive and strategic approach to embrace the age of artificial intelligence (AI). This article argues that China's bold AI practices are part of its broad and incoherent adaptation strategy to governance by digital means. AI is part of a digital technology package that the Chinese authoritarian regime has actively employed not only to improve public service, but also to strengthen its authoritarian governance. China's digital progress benefits from its huge internet market, strong state power and weak civil awareness, making it more competitive than western democratic societies where privacy concern restricts their AI development. However, China's ambitious AI plan contains considerable risks; its overall impact depends on how AI affects major sources of political legitimacy including economic growth, social stability and ideology. China's approach is gambling on its success in (a) delivering a booming AI economy, (b) ensuring a smooth social transformation towards the age of AI and (c) proving ideological superiority of its authoritarian and communist values.
Utilizing state–society relations as a theoretic framework, this study investigates whether the Internet is used to facilitate or impede democratization in authoritarian regimes. The present paper uses governance practice as the mediating mechanism, along with a regime's degree of authoritarianism as a moderator, to measure the direct and indirect effects of the Internet on democratization. This research found that the Internet can be mobilized to suppress political freedom in authoritarian regimes, while at the same time reinforcing the legitimacy of the regime's basis—both of these deter democratization. The Internet, as a tool, must be complemented by democratic deliberation to fully generate democratization.
The People’s Liberation Army (PLA) seeks not only to equal but also to overtake the US military through seizing the initiative in the ongoing Revolution in Military Affairs (RMA). Chinese military leaders believe the form of warfare is changing from today’s ‘informatised’ (信息化) warfare to future ‘intelligentised’ (智能化) warfare. The PLA’s approach to leveraging emerging technologies is likely to differ from parallel American initiatives because of its distinct strategic culture, organisational characteristics, and operational requirements. This research examines the evolution of the PLA’s strategic thinking and concepts of operations, seeking to contribute to the military innovation literature by evaluating major theoretical frameworks for the case of China.
Proliferating military artificial intelligence will leave the world less safe — so we must focus on ethics and global cooperation. Proliferating military artificial intelligence will leave the world less safe — so we must focus on ethics and global cooperation.
By using African countries as laboratories to improve its AI-based technologies, China is also reinforcing African governments’ illiberal and authoritarian tendencies.
The pervasive diffusion of electricity-related technologies at the beginning of the twentieth century has been studied extensively to understand the transformative potential of general purpose technologies (GPTs). Most of what we know, however, has been investigated in relation to the diffusion of their use. This article provides evidence on the county-level economic impact of the technological adoption of electrical and electronic (E&E) technologies in the 1920s in the United States (US). It focuses on measuring the impact of a GPT on technological adopters, i.e., those who are able to develop, transform, and complement it. It is shown that places with patenting activity in E&E technologies grew faster and paid higher wages than others between 1920 and 1930. This analysis required constructing a novel database identifying detailed geographical information for historical patent documents in the US since 1836, as well as developing a text-mining algorithm to identify E&E patents based on patent descriptions.
Nations have recognized the transformational potential of artificial intelligence (AI). Advances in AI will impact all facets of society. A spate of recently released national strategic AI plans provides valuable insights into how nations are considering their future trajectories These strategic plans offer a rich source of evidence to understand national-level strategic actions, both proactive and reactive, in the face of rapid technological innovation. Based on a comprehensive content analysis of thirty-four national strategic plans, this article reports on (1) opportunities for AI to modernize the public sector and enhance industry competitiveness, (2) the role of the public sector in ensuring that the two most critical elements of AI systems, data and algorithms, are managed responsibly, (3) the role of the public sector in the governance of AI systems, and (4) how nations plan to invest in capacity development initiatives to strengthen their AI capabilities.